Original paper
Abstract
Volatility-based filtering is proposed to pre-process historical daily return data of stock indexes before applying to price-based technical analysis trading rules. Any “nearly flat” days which have daily gains or losses less than a threshold about 20% of a daily volatility measure, is filtered out as noise. On six widely followed stock indexes over 1990-2012, the daily return sample sets after filtering retain largely the levels of return mean and skew, and have somewhat higher standard deviation and lower Kurtosis. The filtered sets indeed resemble white noise within the bounds of the daily return filter threshold.
By excluding filtered “nearly flat” days from both the price input and the day counting, three types of well-known technical trading rules are tested, using daily returns of the S&P 500 Index over the past 23 years (1990-2012): two-day runs long/short trading for short term mean reversion; dual moving average cross trend following; and long term channel breakout long/short trading. Filtering improves performance in all tested cases for the whole period, and two sub-periods of lower (1990-1999) and higher (2000-2012) market volatilities. Informational relevance by daily return magnitude, and better alignment with institutional investors or other price-targeted traders are argued as possible reasons for the observed efficacy of daily return filtering for technical trading rule adjustments.
Keywords:Â Daily Return Filter, Technical Trading Rule, Stock Market Index, S&P 500, Runs, Dual Moving Average Cross, Channel Trading
Trading rules
- Target investment: Instruments that mirror the S&P 500 Index (such as ETFs, CFDs, or futures)
- Define “Runs”: Consecutive gaining or losing trading days
- When entering a long position: Commit fully to the SPX when it has demonstrated a decline over a span of n days (where n = 2)
- When going short: Transition to being fully short on the SPX when the index shows a rise for n consecutive days (n being 2 in this case)
- Apply a dynamic filter to exclude fluctuating days by setting a Filter Threshold
- Filter Threshold: d = 20% of SPX daily return standard deviation over last 60-day rolling window.
- Updating the threshold: Every day, recompute the threshold, but leave out days when the absolute daily movement is lower than Filter Threshold.
Python code
Backtrader
import backtrader as bt
class FilteredReversalStrategy(bt.Strategy):
params = (
('n', 2),
('filter_percent', 0.2),
('lookback_period', 60),
)
def __init__(self):
self.dataclose = self.datas[0].close
self.daily_returns = self.dataclose / self.dataclose(-1) - 1
self.runs = 0
def next(self):
std_dev = bt.indicators.StdDev(self.daily_returns, period=self.p.lookback_period)
filter_threshold = self.p.filter_percent * std_dev[0]
if abs(self.daily_returns[0]) < filter_threshold:
return
if self.daily_returns[0] > 0:
self.runs = self.runs + 1 if self.runs > 0 else 1
else:
self.runs = self.runs - 1 if self.runs < 0 else -1
current_position = self.getposition().size
if self.runs <= -self.p.n and current_position <= 0:
self.order_target_percent(target=1)
elif self.runs >= self.p.n and current_position >= 0:
self.order_target_percent(target=-1)
if __name__ == '__main__':
cerebro = bt.Cerebro()
data = bt.feeds.YourSPXDataFeed(dataname='SPX')
cerebro.adddata(data)
cerebro.addstrategy(FilteredReversalStrategy)
cerebro.run()Replace YourSPXDataFeed with the appropriate data feed class for the S&P 500 index tracking instrument (ETF, CFD, or future) you are using.